How to manage organisation context
When you'd do this: when your analyses should follow house rules — a coding standard, a regulatory SOP, a domain glossary, a style guide — and you want REQQA to take those into account automatically rather than pasting them into every prompt by hand.
The Context Library is your organisation's shelf of reference documents. You upload a document once, attach it where it should apply, and from then on REQQA includes it in the AI prompts for the requirements it covers. The documents are shared across every application in your organisation, so one upload can serve several projects.
There is a cost angle, too. Sending the same standards on every analysis would add up, so REQQA uses prompt caching: the shared context is billed in full the first time and at a fraction of the price on every repeat call. You get grounded analyses without paying full freight each time.
Upload a document
- Open Context Library from the Manage menu.
- Click Upload Document.
- Give it a Title — or leave it blank to use the file name — and choose a File. Supported formats are PDF, DOCX, Markdown and plain text.
- Click Upload. REQQA extracts the text, estimates its token size, and opens the document view.
Check the extracted text on the document view before you rely on it. Scanned or heavily formatted PDFs can extract poorly, and poor text in means poor analysis out. If it looks wrong, substitute a cleaner source and upload again.
Attach it where it applies
A document does nothing until you attach it. On the document view you can attach at three tiers:
- Organisation-wide — included for every requirement in the organisation. Use this for rules that always apply, such as a house style guide.
- Template — included only for requirements built from a particular template. Use this for rules that apply to one class of requirement.
- Release — included only for requirements in a particular release. Use this for rules scoped to a single piece of work.
Attach at whichever tiers make sense; a document attached at more than one tier still counts once. To attach at the template or release tier you need an application open, so those pickers know which templates and releases to offer.
Check what a requirement will actually use
From a requirement's detail view, click Context to see its resolved context — the de-duplicated union of everything attached org-wide, to its template, and to the releases it belongs to, in the order REQQA sends them. This is the honest answer to "what will the AI see for this requirement?", and it is where you confirm the right documents are resolving before you run an analysis.
Keep documents current
Uploading a new file under an existing title creates a new revision rather than a separate document. Because that can quietly replace a document other applications rely on, REQQA warns you first: it shows where the existing document is attached — including which application's templates and releases — and lets you Replace it as a new revision or Cancel and rename to keep both.
If a re-upload was a mistake, open the document and use Make current on the good revision in the history. That brings it back as a new active revision and carries its attachments forward. Nothing you do to revisions changes what past analyses recorded — each analysis keeps the exact revision it used.
To take a document out of circulation, use Retire. It disappears from new attachments but stays on record for the analyses that already used it.
Watch the token budget
Context is not free — every attached document adds tokens to the prompt. The document view shows a running total per tier, and the resolved-context page warns you when a single requirement's context goes over the soft budget (20,000 tokens by default). The warning is advisory: it never blocks anything. If a requirement is legitimately large, use Don't warn for this requirement to mute it for that requirement alone — this changes nothing about the context itself and does not touch the requirement's history.
See what caching saves (administrators)
System administrators have two extra views under the Administrator menu:
- Cache & Context Usage — per-organisation cache hit rate, token volumes, and the most-used context documents.
- Token-Savings Experiment — a side-by-side comparison of the same analyses run with no context, with context cache-cold, and with context cache-warm, showing exactly what the caching buys.
Both populate as real analyses run against a caching-capable model, so give them some traffic before reading too much into the figures.